How do voice assistants handle speech recognition and intent parsing?

22 Eki 08:00

I’m trying to understand the core workflow behind modern voice assistants. Specifically, how does the system move from detecting the wake‑word to converting speech into text, then extracting the user’s intent, and finally triggering actions? Which parts typically run on‑device versus in the cloud, and what trade‑offs influence that split? Looking for a high‑level overview of the processing pipeline.